Extended Collaborative Support program provides expert assistance in a wide range of
cyberinfastructure technologies. Any user may request this assistance through the XSEDE allocation process.
The primary goal of this monthly symposium is to allow the over 70 staff members working in ECSS to exchange information about successful techniques used to address challenging science problems. Tutorials on new technologies may also be featured. Two 30-minute, technically-focused talks will be presented each month and will include a brief question and answer period. This series is open to all.
These sessions will be recorded. For this large webinar, only the presenters and host will be broadcasting audio. Attendees may submit questions to the presenters through a moderator by sending a chat message.
March 20, 2018
ECSS Symposium featuring PI Panel
Presenter(s): Michael Cianfrocco (University of Michigan) Cameron Smith (Rensselaer Polytechnic Institute) Jian Tao (Texas A&M University) Sever Tipei (University of Illinois)
Curious about XSEDE's Extended Collaborative Support Services (ECSS)? Join us at our ECSS Symposium webinar on March 20 to hear from a panel of PIs about their experiences working with ECSS! They'll share what it was like requesting ECSS support, what the collaboration was like throughout the course of the project, and how ECSS support helped them achieve results.
Michael Cianfrocco is a Research Assistant Professor at the University of Michigan's Life Sciences Institute. Michael's ECSS project, "Analysis of Cryo-EM data on Comet and Gordon," began with a postdoctoral position with Andres Leschziner's lab at UCSD. Michael has been working with Mona Wong (SDSC) through both ECSS and the Science Gateways Community Institute to develop a gateway that would offer the cryoEM science community a web-based tool to simplify the analysis of data using a standardized workflow running on XSEDE's supercomputers. This gateway will lower the barrier to high performance computing tools and contribute to the fast-growing field of structural biology.
Cameron Smith is a Computational Scientist at the Scientific Computation Research Center at Rensselaer Polytechnic Institute. Cameron's project, "Adaptive Finite-element Simulations of Complex Industrial Flow Problems" focuses on scaling and performance analysis of adaptive in-memory workflows using PHASTA CFD, EnGPar load balancing, and PUMI unstructured mesh services on Stampede2's Knights Landing processors. The workflows are executed through the PHASTA science gateway. Cameron worked with ECSS staff Lars Koersterke and Lei Huang (both at TACC) on this project.
Jian Tao is a Research Scientist in the Strategic Initiatives Group at Texas A&M Engineering Experiment Station and High Performance Research Computing at Texas A&M University. Jian's work, "Deploying Containerized Coastal Model on XSEDE Resources," first began while he was at Louisiana State University. The goal is to develop and deploy enhancements into the SIMULOCEAN science gateway, integrating new Docker features of Bridges and Globus capabilities for authentication, file transfer and sharing. The PI worked with Mona Wong and Andrea Zonca (SDSC) and Stuart Martin from the Globus team.
Sever Tipei is a Professor of Composition-Theory in the School of Music at University of Illinois' College of Fine and Applied Arts. His project, "DISSCO, a Digital Instrument for Sound Synthesis and Composition" involves optimization and parallelization of the multi-threaded code DISSCO (developed jointly at the UIUC Computer Music Project and at Argonne National Laboratory). DISSCO combines the field of Computer-assisted Composition with that of the Sound Design in a seamless process. Sever has worked with ECSS staff Paul Rodriguez and Bob Sinkovits (both at SDSC) on this project.
February 20, 2018
Deep Learning: An Increasingly Common HPC Task
Presenter(s): Paola Buitrago (PSC) Joel Welling (PSC)
Presentation Slides Joel Welling Slides
Presentation Slides Paola Buitrago Slides
Deep learning is a highly compute- and data-intensive category of tasks with wide applicability in science as well as industry. Join Paola Buitrago and Joel Welling from the Pittsburgh Supercomputing Center in two talks that will provide an overview of the current deep learning landscape and examples of the deep learning environments available to XSEDE users. Paola will provide a brief history of the field and an update on its technical performance, with examples from domains as diverse as vision and theorem proving. Joel will follow with a description of the PSC's support for two major deep learning packages, TensorFlow and Caffe.
January 16, 2018
An Introduction to Jetstream
Presenter(s): Virginia Trueheart (TACC)
Jetstream is an interactive computing resource designed to make High Performance Computing accessible to users that are not part of traditional HPC fields. This tutorial aims to introduce Jetstream's capabilities to this expanded user base. It will demonstrate how to access the system, make use of the various Virtual Machines available, and use publicly available images to assist with research. It will also cover how to create, modify, and save personal images that can be customized to individual workflows and be saved long term for reference in publication.
Visualizations of Simulated Supercell Storm Data
Presenter(s): Greg Foss (TACC)
Principal Investigator(s): Amy McGovern (University of Oklahoma)
XSEDE ECSS project: High Performance Computing Resources in Support of Spatiotemporal Relational Data Mining for Anticipation of Severe Weather.
Amy McGovern and her collaborator Corey Potvin from the National Severe Storms Laboratory are developing and applying novel spatiotemporal data mining techniques to supercell thunderstorm simulations, with the goal of identifying tornado precursors. The overall goal of the project is to improve tornado warning lead time and accuracy by integrating into the "Warn on Forecast" project, a National Oceanic and Atmospheric Administration research program tasked to increase tornado, severe thunderstorm, and flash flood warning lead times.
XSEDE ECSS staff was enlisted to see what could be found using 3D visualization techniques and an interactive user interface. The resulting images and animations will assist in defining storm features and as input to the data mining: ensuring automatically extracted objects match visually identified ones. This talk will feature graphics from a selection of three (5.7 TB) datasets, with visualization samples identifying various supercell thunderstorm features.
December 19, 2017
Jupyter Notebooks deployments at scale for Gateways and Workshops
Presenter(s): Andrea Zonca (SDSC)
Andrea Zonca (SDSC) will give an overview on deployment options for Jupyter Notebooks at scale on XSEDE resources. They are all based on deploying Jupyterhub on Jetstream, then either spawn Notebooks on a traditional HPC system or setup a distributed scalable system on Jetstream instances either via Docker Swarm or Kubernetes.
Deployment and benchmarking of RDMA Hadoop, Spark, and HBase on SDSC Comet
Presenter(s): Mahidhar Tatineni (SDSC)
Data-intensive computing middleware (such as Hadoop, Spark) can potentially benefit greatly from the hardware already designed for high performance and scalability with advanced processor technology, large memory/core, and high performance storage/filesystems. Mahidhar Tatineni (SDSC) will give an overview of the deployment and performance of Remote Direct Memory Access (RDMA) Hadoop, Spark, and HBase middleware on the XSEDE Comet HPC resource. These packages have been developed by Dr. D.K. Panda's Network-Based Computing (NBC) Laboratory at the Ohio State University. The talk will cover details of the integration with the HPC scheduling framework, the design and components of the packages, and the performance benefits of the design. Applications tested include the Kira toolkit (astronomy image processing), latent Dirichlet allocation (LDA) for topic modeling, and BigDL (distributed deep learning library).
October 17, 2017
Geodynamo Simulation Code for Paleomagnetic Observations
Presenter(s): Shiquan Su (NCAR) Chad Burdyshaw (NICS)
Principal Investigator(s): David Gubbins (Scripps Institution of Oceanography at UCSD)
This study characterizes a geodynamo simulation code for paleomagnetic observations targeted to run on TACC Stampede2 KNL cluster; a hybrid, distributed many-core parallel architecture. Issues examined are parallel scaling across distributed nodes and within the many core architecture, as well as vectorization efficiency, arithmetic intensity and memory throughput.
This presentation includes two parts. In the first part, Shiquan Su from NCAR will introduce the background of the project, the parallelization algorithm, the experience on Stampede KNL cluster, OpenMP treatment, and the two approaches to run the project jobs on the machines. In the second part, Chad Burdyshaw from UTK takes a close look at the code. Chad will discuss the tools used to interrogate performance, observations, remedies, and potential solutions.
September 19, 2017
COSMIC2 - A Science Gateway for Cryo-Electron Microscopy with Globus for Terabyte-sized Dataset
Presenter(s): Mona Wong-Barnum (SDSC)
Principal Investigator(s): Andres Leschziner (UCSD) Michael Cianfrocco (University of Michigan)
Structural biology is in the midst of a revolution. Instrumentation and software improvements have allowed for the full realization of cryo-electron microscopy (cryo-EM) as a tool capable of determining atomic structures of protein and macromolecular samples. These advances open the door to solving new structures that were previously unattainable, which will soon make cryo-EM a ubiquitous tool for structural biology worldwide, serving both academic and commercial purposes. However, despite its power, new users to cryo-EM face significant obstacles. One major barrier consists of the handling of large datasets (10+ terabytes), where new cryo-EM users must learn how to interface with the Linux command line while also dealing with managing and submitting jobs to high performance computing resources. To address this barrier, we are developing the COSMIC2 Science Gateway as an easy, web-based, science gateway to simplify cryo-EM data analysis using a standardized workflow. Specifically, we have adapted the successful and mature Cyberinfrastructure for Phylogenetic Research (CIPRES) Workbench  and integrated Globus Auth  and Globus Transfer  to enable federated user identity management and large dataset transfers to Extreme Science and Engineering Discovery Environment's (XSEDE)  high performance computing (HPC) systems. With the support of XSEDE's Extended Collaborative Support Services (ECSS)  and the Science Gateway Community Institute's (SGCI) Extended Developer Support (EDS), this gateway will lower the barrier to high performance computing tools and facilitate the growth of cryo-EM to become a routine tool for structural biology. Talk previously given at PEARC'17
First steps in optimising Cosmos++: A C++ MPI code for simulating black holes
Presenter(s): Damon McDougall (ICES)
Principal Investigator(s): Patrick C. Fragile (College of Charleston)
This ECSS project is to have Cosmos++ run on Stampede2 effectively. Stampede2, at present, is made up entirely of Intel Xeon Phi nodes. These are low clock-frequency but high core-count nodes, and there are some challenges associated with running on this hardware efficiently. Although the project's end goal is to hybridise a pure MPI code, this talk will focus on some of the initial steps we have taken to improve serial performance and how these steps relate to C++ software design. Prior knowledge of compiled languages and custom types would be beneficial but isn't required.
August 15, 2017
HTC with a Sprinkle of HPC: Finding Gravitational Waves with LIGO
Presenter(s): Lars Koesterke (TACC)
Principal Investigator(s): Duncan Brown (Syracuse University) Josh Willis (Abilene Christian University)
XSEDE is supporting the LIGO project to detect signatures of gravitational waves in a stream of data generated by (currently) two observatories in the U.S., located in Washington State and Louisiana. I will report on an ECSS project tasked to improve the performance of one of the largest (most resource demanding) pipelines called pycbc (python compact binary collision). The software evolved from a slow and performance-unaware state to a high-performing pipeline capable of utilizing Xeon, Xeon Phi, and Nvidia GPU architectures alike. Achieving high performance required only a few sprinkles of HPC (High Performance Computing) on top of a HTC (High Throughput Computing) pipeline. While the HPC pieces relevant for this particular project are all well known to ECSS staff it may be surprising what was missing in the considerations of the software developers. Hence this is more a story of how to educate users than a story of new and groundbreaking HPC concepts. Nevertheless I am confident that my fellow ECSS staffers will find this project interesting and enlightening.
Enabling multi-events 3D simulations for earthquake hazard assessment
Presenter(s): Yifeng Cui (SDSC)
Principal Investigator(s): Morgan Moschetti (USGS)
Researchers from USGS use Stampede to perform a series of computationally intensive simulations for improved understanding of earthquake hazards. Hercules, a finite element solver developed at CMU, is used to make the calculations which combines meshing, partitioning and solving functions in a single, self-contained code. Meshing employs a highly efficient octree-based algorithm that scales well. The simulation results are used to investigate the effects of complex geologic structure and topography on seismic wave propagation and ground-shaking hazards, and to evaluate model uncertainties in U.S. seismic hazard models. This talk will provide an overview of current status of the seismic hazard analysis research, and introduce the code performance, the optimizations involved in supporting multi-event simulations for this study through the ECSS project.
June 16, 2015
A Short Story of Efficiently Using Two Open-Source Applications on Stampede
Presenter(s): Ritu Arora (TACC)
This presentation will cover a summary of two challenges and solutions related to running the DROID (Digital Record Object Identification) and the FLASH astrophysics code on a large number of nodes on Stampede.
DROID is a software tool developed by The National Archives to perform automated batch identification of file formats. It is written in Java and works well when only one copy of it is run on a node. PI Jesscia Trelogan from the Institute of Classical Archaeology at UT Austin has been using DROID as part of her workflow for managing a large archaeological data collection. It would take her more than 2 days to extract metadata from about 4.3 TB of data using DROID on a local server. Since the process of culling and reorganizing the data collection is iterative, the metadata extraction using DROID needs to be done often. The goal of the ECSS project with PI Trelogan was to provide support in leveraging Stampede for parts of her workflow, which includes DROID, so that the overall time-taken in conducting all the steps in the workflow is reduced. The main challenge in using DROID on Stampede was related to executing its multiple copies in parallel on different nodes in a batch mode. An overview of this challenge and its solution strategy will be discussed during this presentation.
In another project, a copy of the FLASH astrophysics code was optimized such that the code does striped I/O on the Lustre File System. This project was proposed after it was found that a user overloaded the Lustre servers (which eventually became unresponsive) while running FLASH on 7000+ cores. The problem was related to the step that involved reading a checkpoint file. An overview of the problem and its solution will be included in this talk.
Optimization of Text Processing for the WordFlare Knowledge Graph
Presenter(s): Robert Sinkovits (SDSC)
Principal Investigator(s): Michael Douma (IDEA)
The goal of the WordFlare project is to create a tablet-based app to engage K-12 and lifelong learners in exploring language and knowledge. The app is based on a massive thesaurus and features dynamic visualizations of word relationships. Approximately 9% of the content is human-curated, while the other 91% is derived using computational methods executed on XSEDE resources. In this talk, I will describe the steps taken to accelerate two key steps in the automated text processing – optimization of the Latent Dirichlet Allocation (LDA) algorithm and the development of a fast method to simultaneously search for large numbers of words in a corpus. The speedups we obtain are highly problem dependent, ranging from 1.5-2.2x for the LDA algorithm and up to 1500x for the word search when using a large reference dictionary (e.g. the 400K words found in Wiktionary).
May 19, 2015
ECSS experience with non-traditional HPC users
Presenter(s): Junqi Yin (NICS)
Principal Investigator(s): Annette Engel (U. Tenn) Yong Zeng (UMKC)
Mothur is an open source bioinformatics pipeline used for biological sequence analysis that has gained increasing attention in the microbial ecology community. Because a large set of functionalities in Mothur are memory bound, it is well suited for shared memory architectures. I will discuss performance results for several commands in Mothur that are popular in the operational taxonomic unit analysis, and show that pipeline processes can be accelerated by orders of magnitude faster.
Real-time Bayesian estimation for financial ultra-high frequency data is plagued with the curse of high dimensionality. Methods have been developed to manage this problem through the use of MPI. By porting to CUDA, I'll show that an adequately equipped GPU workstation can rise to the task, producing reasonably real-time results with actual data from financial markets.
P3DFFT: a scalable open-source solution for Fourier Transforms and other algorithms in three dimensions
Presenter(s): Dmitry Pekurovsky (SDSC)
P3DFFT is an open-source package developed at SDSC. It implements three-dimensional Fourier Transforms and other algorithms, in a highly scalable and efficient way. P3DFFT achieves good scaling on hundreds of thousands of compute cores. It has received much interest and use from scientists in diverse fields such as DNS turbulence simulations, astrophysics, oceanography and material science. Recently it has been the subject of an internal ECSS project, aimed at making it XSEDE community software. It has been ported, tested and documented on the largest computational systems at XSEDE. Additional features have been added to help widen the impact in the community. In this presentation I will go over the main features of P3DFFT, including the recently added, and review how users of XSEDE can access it on XSEDE platforms.
April 21, 2015
reproducibility@XSEDE: Reporting Back to our Colleagues
Presenter(s): Doug James (TACC) Carlos Rosales (TACC) Nancy Wilkins-Diehr (SDSC)
The reproducibility@XSEDE workshop (www.xsede.org/reproducibility) was a full-day event held in conjunction with XSEDE14. The workshop featured an interactive, open ended, discussion-oriented agenda focused on reproducibility in large-scale computational science. This presentation includes (1) independent reactions to the event by three of the workshop principals; and (2) an open discussion on the topic of reproducibility in general.